- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 1
Read review
Paper summary
This work discusses Do‑CJEPA, a modification of Causal‑JEPA (C‑JEPA) for object‑centric world modeling under distribution shift. Given frozen SAVi slots matched to object roles, it (i) restricts transformer attention for each target slot to its lag‑one parents in a slot‑level ADMG (or to possible parents of a PAG estimated with G‑CaRL + LPCMCI when no graph is given), and (ii) adds a stop‑gradient regression loss toward the mean of the graph‑surgery estimand , , estimated by ID‑GEN from source data only. The claim is that this yields predictors invariant to mechanism shifts in the presence of latent confounders, without knowing the shift target or time.
Critical strengths
-
The evaluation protocol is structurally leak‑free and the intervention‑aligned analysis is a reusable contribution in its own right. The predictor does not receive , the intervened role, or the intervention value. ID‑GEN, the causal‑discovery pipeline, graph aggregation, and checkpoint selection all use source trajectories only. The alignment in 4.2, with the directly intervened token removed only at , is the right way to read a one‑time intervention: it separates the unavoidable error at the shifted node from error that propagates to nodes whose mechanisms are unchanged, which is the only quantity an invariance method can legitimately be judged on.
-
The paper draws the correct conceptual line between predictive sufficiency and invariance, and respects it in the PAG variant. 2.2 correctly states that an influence neighborhood is a Bayes‑optimality object, not a causal one: a token can be in precisely because it is a child of the selection variable. This is a sharper reading of C‑JEPA than C‑JEPA's own authors give. The same discipline shows up in 3 and footnote 4: a PAG is read only as a possible‑parent restriction ( and excluded), never promoted to an ADMG from which an identification functional would be computed.
Critical weaknesses
-
By the paper's own Eq. 6, the interventional estimand equals the conditional, so the "do" contributes nothing, and the mechanism the main text credits for the gain is contradicted by the method as described. In Sec. 4 the authors write that in Interventional Pong , hence . This means (a) regresses the predictor toward the same conditional mean that already converges to under the same input restriction, i.e. it is self‑distillation from an 8‑component GMM; (b) the ID/IDC machinery of Sec. 2.1, Fig. 2 and the conditioning set of Def. 2.1 are never exercised; and (c) the "known ADMG" gives every target all three lag‑one slots, so the causal restriction is exactly "attend only to frame " (the Do‑JEPA bullet states each batch observes only frame ). Nothing object‑specific is removed. Sec. 4.1 then explains the improvement as "target‑specific attention prevents invariant predictions from depending on the intervened right‑paddle slot," but the mask includes for every target; what it removes is older frames.
-
The headline gains come mostly from how the baseline was trained, and the paper's own figures show this. C‑JEPA is trained for a single task, while the Do‑variants are trained to predict frame from the preceding frames for every , and all models are then evaluated at every . The baseline is tested on prediction lengths it ddid not see, and Fig. 4 shows the cost: on frames before the intervention (), where nothing has shifted yet, C‑JEPA already sits near while the Do‑variants sit near , a gap on unshifted data that equals C‑JEPA's reported OOD gap in Fig. 3. So the abstract's / reductions and " smaller OOD gap" are largely measuring the baseline's handicap. What remains is further weakened by two facts: Sec. 4 describes the shift as a one‑time nudge to the right paddle after which the ordinary dynamics resume, to which any correct one‑step Markov model is invariant by construction with no causal knowledge; and in Fig. 5 the same‑slot OC‑DINO‑WM curve crosses below every Do‑variant and finishes lowest at .
Questions for authors
-
Could you report matched‑boundary C‑JEPA with a context window of exactly 1 frame (and 2 frames), without any graph. If this reproduces the mask‑only row, what remains of the causal contribution in the ADMG variant?
-
When is the do‑target different from the conditional? Eq. 6 gives in the benchmark, and Sec. 3 falls back to whenever is confounded with a lag‑one predictor. Could you please give one slot‑level graph in which the target is both identifiable and , and (b) evaluate on a setting that instantiates it.
-
Is there a hidden confounder? Could you report a linear probe from a single matched SAVi slot to simulator ball velocity? If it is accurate, Fig. 1b's bidirected edges are not the correct latent projection, and the success of a one‑frame‑context predictor is explained.